On Thu, 19 Dec 2024 08:05:17 +0200 Matti Vaittinen <mazziesaccount@xxxxxxxxx> wrote: > On 18/12/2024 22:53, David Lechner wrote: > > On 12/18/24 2:38 AM, Matti Vaittinen wrote: > >> Hi dee Ho peeps, > >> > >> I started drafting a driver for a ROHM DAC. I took a quick look at the ad5624r_spi.c, and the use of the 'scan_type' -field in the struct iio_chan_spec puzzled me. > >> > >> I think this field is used by the driver to convert the data from user to register format while performing the INDIO_DIRECT_MODE raw writes. I don't spot any buffer usage. Furthermore, as far as I can say the 'sign' and 'storagebits' are unused. > >> > >> My understanding has been that the scan_type is only intended for parsing the buffered values, and usually when the data direction is from driver to user. > >> > >> I suppose I shouldn't copy the ad5624r_spi.c use of scan_type to a new driver. I'm somewhat tempted to send a patch which drops the scan_type from the ad5624r_spi.c, and adds the 'realbits' and 'shift' to the driver's internal struct ad5624r_state. This, however, will change the interface to userland so maybe it's best to not do that. > > I think I was wrong here. I suppose plain scan_type population does not > result user visible entries if buffer is not created. So, confusion > stays in driver - but it also means changes wouldn't impact the userland. > > >> > >> I wonder if I am missing something? (That wouldn't be unheard of XD). If not, then at least a documentary patch with a comment "don't do this in new drivers" might be Ok, or how do you see this? > >> > >> Yours, > >> -- Matti > >> > > > > I think scan_type is a convenient place to store this information even if > > buffers aren't implemented. The struct is there whether we use it or not, > > Valid point. > > > so > > might as well use it. And if buffer support is ever added, that is one less > > thing to do (removing the duplicate fields). > > I find populating the scan_type still somewhat confusing for a reader. > Kinda willing to hear what Jonathan thinks of it, he probably has > broadest view on how to keep things consistent in IIO. If it is usual to > use the scan_type without buffer, then this is totally fine with me. I'm against filling it in if there is no use being made of it (which people sometimes do) but I don't mind it being used for this sort of purpose. The reasoning being that we wouldn't want duplication if buffered mode was supported, and this is just using it in the way we would use it under those circumstances. I'm also against carrying it around if it's actually a constant, so bring this in only when dealing with a driver supporting multiple values (either multiple devices or channels where this is the only difference). > > I suppose the shifts and amount of bits are constants? In that regard > one could also just use a define, which would make it possible to not > add this information to any of the structs. Absolutely. If there is only one value used and no buffered support I'd prefer to see it as a define used inline in the code. The usual principle of don't generalize until you need to applies here. > > Out of the curiosity, do we use 'input buffers' in IIO? This far I've > mostly worked with IIO devices focusing on output. There are some. Look for IIO_BUFFER_DIRECTION_OUT. Eg. dac/ad3552r.c It's relatively new infrastructure and for many DACs we support it isn't worth the effort as they are more about tuning a signal than waveform generation etc. Hence, not that many drivers yet and some of those are using the IIO backend stuff with fpga IPs to get the data rates. Jonathan > > Thanks for sharing your opinion! > > Yours, > -- Matti >
